About & methodology
How this list is made
HackerPods reads Hacker News comments about podcasts and keeps the shows that different people recommend again and again, over years rather than in one viral thread.
This edition: 440 shows · 18 episodes · 12,599 labeled mentions · Data as of
The bar
The short version: three people, three months. A show is listed when at least three different people recommend it, in at least three different calendar months. Only recommendations count. A passing reference, a joke, or a criticism stays in the data and out of the score. A host promoting their own show is kept and never scores.
The score
score = months with a recommendation + 0.5 × months with a strong recommendation
Counting months rather than comments means one enthusiastic thread cannot carry a show. A show recommended steadily for a decade outranks one that was briefly everywhere. Controversial shows are ranked by the same rule as everything else; nothing is removed by hand.
How comments are read
We start from roughly 70,000 comments that mention podcasts or link to a podcast app. Rules drop comments that name no show. A language model then reads the rest in batches, comments with podcast links and recommendation words first (this edition covers those; plainer mentions are queued for a later refresh), and records each show named, whether the comment recommends it, and a short verbatim excerpt. Show names are matched to the Apple Podcasts directory so spellings merge into one entry.
How accurate the labels are
Before the full run, two language models from different vendors labeled the same stratified sample of 200 comments. They agreed outright on 65.5% of them; every disagreement was then re-read and adjudicated against the comment itself. Measured against that adjudicated sample, the labeler used for this edition scored:
| Measure | Result | Bar |
|---|---|---|
| Show-name precision | 96.9% | 90% |
| Show-name recall | 96.3% | 80% |
| Recommendation precision | 91.9% | 85% |
Caveat: rows where both models agreed were taken as correct without a second reading, which flatters both. Treat the figures as a floor check, not a guarantee.
Listening
Play buttons stream each episode directly from the show’s own public RSS feed, the same file any podcast app reads. HackerPods hosts no audio and does not proxy it. Recommended episodes are matched to feed items by title; when an episode has left the public feed, the page says so and links to the show instead.
Sources and credit
Every excerpt is the commenter’s own wording and links to the original comment on Hacker News. The credit belongs to the people who took the time to write them. HN data via ClickHouse SQL Playground, a mirror of the official HN Firebase API. Artwork belongs to each show and comes from the Apple Podcasts directory.
Take it with you
The data (shows, episodes, every labeled mention; ODC-By 1.0) · Download the zip · DATA.md · OPML of every show · Episode RSS
Analytics: Cloudflare Web Analytics, which sets no cookies and does not track individuals. Saved shows stay in your browser.